Anthropic's new AI model can create C compiler from scratch
What's the story
Anthropic, the AI start-up that recently made headlines with Claude Cowork, has unveiled its latest model, Claude Opus 4.6. The new model packs higher coding intelligence, longer reasoning memory, and an "agent teams" feature that lets multiple AI agents work on one project simultaneously. In a recent experiment, 16 of these agents were able to build a C compiler from scratch in just two weeks. The new release marks a major milestone in Anthropic's product development journey and intensifies competition.
Performance review
The compiler can build bootable Linux on multiple architectures
The C compiler created by the agents can build a bootable Linux 6.9 on x86, ARM, and RISC-V architectures. However, it does have some limitations such as missing a 16-bit x86 backend for Linux booting and being less efficient than established compilers. The project took around $20,000 in API costs and was completed without any internet access through nearly 2,000 sessions of work by the AI agents.
Global reaction
Each Claude agent work in its own isolated container
The announcement of this groundbreaking experiment has drawn a flurry of reactions on social media. Derya Unutmaz, a professor at The Jackson Laboratory, was left astonished by the feat. The project was led by Nicholas Carlini, a researcher on Anthropic's Safeguards team, and saw each Claude agent work in its own isolated container with a simple synchronization mechanism for efficient division of labor and conflict resolution among agents.
Testing success
The compiler passed the GCC torture test suite
The C compiler developed by the Claude agents was put to the test against challenging benchmarks. It passed 99% of the GCC torture test suite and even compiled and ran the classic game Doom. However, it should be noted that this program is a highly complex task, one that even experienced programmers may struggle with when creating a C compiler from scratch.
Model capabilities
Aim to streamline complex workplace tasks
Claude Opus 4.6 is said to be a major leap forward in how AI tackles complex workplace tasks. The company highlighted its potential for financial modeling that combines complex regulatory filings and market data, as well as document and presentation outputs needing minimal refinement. "Claude Opus 4.6 gets much closer to production-ready quality on the first try than what we've seen with any model," Anthropic said, noting deliverables will require "less back-and-forth" to finalize.
Business strategy
Anthropic remains committed to ad-free experience
Unlike OpenAI, which has started showing ads to its free-tier users, Anthropic has vowed to keep its Claude chatbot ad-free. The company believes that advertisements are "incongruous" with the personal nature of user conversations. Instead of relying on ad revenue, Anthropic focuses oenterprise deals and paid subscriptions for income. This difference in approach is highlighted in its first n Super Bowl ad campaign scheduled for this weekend.